Constructing Simplified Models of Transient Dynamics in Power Systems

Power production and distribution play an essential role in modern society. The power grid is a complex network of components whose state is often perturbed in response to varying loads, short-circuits, and other disturbances. Effective control requires parsimonious models which make accurate predictions in real-time. Power system models have a number of features that makes their improvement challenging--they are multi-level, multi-user and multi-physics. They are nonlinear and time varying, and discrete structures, such as graphs, are strongly blended with continuous dynamics. We explore the use of model reduction methods to find simplified representations of power systems. Our goal is to identify the emergent structures that govern the large-scale dynamics of the grid. Our method is based on information geometry, and uses advances in computational differential geometry to characterize high-dimensional manifolds in the space of measurements. We present results using a 14-bus test system with 58 parameters.